Skip to content
Benchmaker
Applying the framework

Practice

Workflows, design moves and audit tools.

  1. 1The One SpineInteractiveTen steps from claim to frame review that organize every procedure in this handbook, in the order that keeps you out of trouble.10 min read
  2. 2Verifiability engineeringInteractiveDon't ask whether a domain is verifiable. Decompose it, make artifacts emit evidence, compile what you can, and keep the residue honest.10 min read
  3. 3Guarantee before you estimateInteractiveMake bad states unreachable, verify the scaffold you control, and spend samples only on what construction can't reach.9 min read
  4. 4Inverse evaluation designTreat expert accept/reject decisions as evidence about a latent judgment, and use the posterior to find criteria, thresholds and the questions worth asking.10 min read
  5. 5Auditing an evalInteractiveSeven questions if you have an hour, twenty-five if you have a week: audit someone else's eval, or your own.10 min read
  6. 6Anti-patterns and failure vocabularySeventeen named ways evaluator systems fail, each with its tell, its layer and its fix.14 min read
Start with The One Spine